We assess the construct and predictive validity of cognitive and socio-emotional skills in Pakistan using two innovations in measurement and sampling. First, we developed and implemented a battery of tests to capture cognitive and socio-emotional skills among young adults. We measured socio-emotional skills using both self-reported and task-based instruments and psychometrically verified the validity of the different components. For cognitive skills, we measured standard literacy and numeracy as well as skills useful for everyday life. We demonstrate the reliability and construct validity of these measures compared to previous attempts in the literature. Second, we constructed a panel that follows respondents from their original rural locations in 2003 to their locations in 2018, a period over which 38% of respondents left their native villages. We show that the predictive validity of our skills measures is mediated by the migration decision. Among male migrants, labor earnings are strongly correlated with years of schooling, but not socio-emotional skills. Among male non-migrants, wages are associated with socio-emotional skills, but not years of schooling. These associations are consistent with similar data from rural Cambodia, a region with similar levels of schooling but different patterns of migration and labor force participation.